Assessing SES differences in life expectancy: Issues in using longitudinal data

1. Elsie Pamuk, Kim Lochner, Nat Schenker, Van Parsons, Ellen Kramarow National Center for Health Statistics Centers for Disease Control and Prevention Assessing SES differences in life expectancy: Issues in using longitudinal data

2. Why is this important? • Socioeconomic disparities • Life expectancy • Longitudinal data

3. Socioeconomic disparities • Focus of health policy “Inequalities in income and education underlie many health disparities in the United States.” Healthy People 2010: Understanding and Improving Health

4. Life Expectancy • Useful (and intuitive) measure for summarizing mortality rates across all ages • Derived from a life table

7. Longitudinal data • Allows the calculation of life expectancies for groups defined by survey characteristics • Eliminates numerator/denominator inconsistencies

8. “This result tends to validate a specific concern about U.S. mortality estimates calculated from death certificate data. NCHS typically publishes U.S. mortality rates by education level for ages 25–64 because of concerns about the accuracy of death certificate education information at older ages….” Limits the age range & number of education groups that can be compared

9. Issues arising from using longitudinal data to estimate life expectancies

10. Data quality

11. How Records are Linked NCHS Records NCHS Records Administrative Records Administrative Records SSN SSN SSN Name Name Name Name DoB DoB DoB DoB Sex Sex Sex Sex State ofBirth State of Birth State of Birth State of Birth Race Race Race Race State of Residence State of Residence State of Residence State of Residence MaritalStatus Marital Status Marital Status Marital Status Scoring system, Scoring system, Potential matches Potential matches Non matches Non matches clerical review clerical review True matches True matches Non matches Non matches Linked Data File Linked Data File

12. Probabilistic Matching Procedure • Missing identifying information from survey respondent  ineligible for matching • Ineligibility not random across groups

13. Percent of survey participants ineligible for NDI match: NHIS 1986-2004 survey years

14. Addressing insufficient information for matching: • Ineligibility-adjusted weights • Reweighting of matched respondents to be representative of civilian, non-institutionalized population • Exclusion of problem groups • No separate analysis of Hispanics

15. Generating appropriate measures of sampling variability

18. Obtaining standard errors for life expectancy derived from longitudinal data Ideally, should take into account: • Correlation within age-groups resulting from survey sampling design • Correlation across age-groups resulting from respondents contributing to more than one age group

19. Case study of the sensitivity of life expectancy standard errors to study and sample design: Compared standard errors derived by • Chiang method (traditional) • Balanced Repeated Replication Hybrid methods: • BRR & Chiang • Taylor Series (SUDAAN proc RATIO) & Chiang

20. Comparison of standard errors

21. Study Conclusions • Traditional method (Chiang) consistently underestimate standard errors If balanced repeated replication procedure is impractical, • Taylor Series (SUDAAN proc Ratio)/Chiang hybrid can yield reasonably accurate results for finer subgroups

22. Exclusion of the institutionalized population

24. Examination of the effect of excluding the facility dwelling elderly: • Used MCBS data for facility dwelling beneficiaries for 1992-96/98 and 2000-2004/06 • Calculated death rates by sex and education level for ages 70-89 and combined with NHIS/NDI rates

26. Using longitudinal data to examine SES differences in life expectancy • Adds to our ability to routinely monitor SES differences in mortality • Brings with it several methodological challenges